Model predictive control by differential neural networks approach

In this paper a new model predictive neural control is suggested. It consists of the application of the model predictive method to control a nonlinear uncertain system where the information is reduced. The uncertain plant was approximated by a special class of dynamic neural network observer (projectional observer) that uses some sort of information regarding the set where the states remain. A novel method leads to construct an approximate model of the uncertain system where the controllability condition is ensured. The model predictive control was designed using the information obtained by the proposed observer. The upper bound for the tracking error was established if the controller is applied. Simulation regarding the control of a biotechnological process is carried out.